Zheng 2005.

Methods	RCT (n = 33) Allocation concealment not described
Participants	Intubated participants with COPD with severe respiratory failure due to pulmonary infection, who were capable of self care in the past year
Interventions	Once pulmonary infection had been significantly controlled (PIC window appeared), participants were randomly assigned to NPPV versus invasive PS. NPPV was administered in BiPAP mode with a face mask. The criteria for the PIC window were (1) chest x‐ray showed improvement in infectious infiltrates, (2) WBC count was less than 10 × 109/L, (3) sputum was decreased and was less purulent or white, with sputum tenacity decreased (lower than grade II). In the IPPV group, participants were weaned with PS. Inspiratory pressure was decreased gradually to less than or equal to 8 cm H2O to keep RR less than 28 breaths/min, VT approximately 8 mL/kg, SpO2 greater than 90% and PaCO2 between 45 and 60 mm Hg or at baseline levels. If participants were stable for four hours with a spontaneous cough, they were extubated. In the NPPV group, participants were ventilated with a nasal or oral mask using BiPAP. Inspiratory pressure and FiO2 were adjusted to keep RR lower than 28 breaths /min, VT approximately 8 mL/kg, SpO2 greater than 90% and PaCO2 between 45 and 60 mm Hg or at pre‐extubation levels. All participants received 4 to 6 cm H2O PEEP to reduce the work of breathing from intrinsic PEEP. The duration of NPPV was longer than two hours initially, and investigators gradually decreased the duration of NPPV and inspiratory pressure daily until NPPV was required for less than two hours per day and inspiratory pressure was less than 10 cm H2O
Outcomes	1. Hospital mortality 2. Total duration of MV 3. ICU LOS 4. Hospital LOS 5. VAP 6. Time to PIC window 7. Duration of ETMV
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Prospective, randomized, controlled clinical study. No mention of sequence generation
Allocation concealment (selection bias)	Unclear risk	Not reported
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	No denominators reported in binary outcomes
Selective reporting (reporting bias)	Low risk	Clinically important outcomes reported